Robostral Navigate
2026-07-08 · Publisher : Mistral AI · Domain : Artificial Intelligence
Mistral AI introduces Robostral Navigate, its first embodied navigation model: 8 billion parameters, a single RGB camera, trained in simulation, to guide robots from natural-language instructions.
What changes
- Navigation from RGB images and a natural-language instruction, with no depth sensor, LiDAR or multiple cameras.
- 76.6% on R2R-CE (validation on unseen environments), ahead of the best single-camera system by 9.7 points and the best multi-sensor system by 4.5 points.
- Navigation by pointing, with online reinforcement learning (CISPO algorithm).
- Generalization announced to wheeled, legged and flying robots.
### Specifications - Parameters: 8B. Input: RGB images and text. Output: navigation commands. - License, pricing and public API: not disclosed; access through commercial contact ("talk with our team"). - Technical report: arXiv 2607.20785 (referenced on the page). - Context and cutoff: not disclosed. ### Pricing Pricing not disclosed at launch. ### Benchmarks | Benchmark (R2R-CE) | Success | |---|---| | Validation seen | 79.4% | | Validation unseen | 76.6% | Scores published by the vendor. ### Key takeaways A specialized model for robotics opens up industrial uses for Mistral (logistics, manufacturing), but access remains on request, with no public weights or pricing.
Links
Tags : LLM, Mistral, Robostral, Robotique, Vision